We thank Dr. Zhu et al. for their valuable inputs1 regarding our study.2 Dr. Zhu rightly points out that there is a relatively large proportion of zeros in our data set, which may support the use of mixed zero-inflated Poisson models. In our case, however, the use of such statistical approach was not formally indicated: the rationale underlying zero-inflation is that certain individuals might—because of some unknown factors—not be able to present values other than zero. If this is indeed the case, the logistic regression part of zero-inflated models is used to model that inability. In our trial, though, in-depth review of our longitudinal data revealed that most patients presented values other than zero on at least one of the time points. Most notably, the percentage of patients presenting zero values over the entire observation time was only around 6%. This percentage was deemed too low to justify the use of a mixed zero-inflated Poisson model.
Nonetheless, because the comparison of varying methodologic approaches is always valuable in the appraisal of results robustness, a mixed zero-inflated Poisson model was fitted to our main outcome (i.e., counts of circulating tumor cells over time). Our findings remained unchanged (i.e., there was no evidence of a difference between sevoflurane and propofol anesthesia).
A second concern of Dr. Zhu is that some variables that may affect the outcome were not captured by our analysis. It is worth remembering, however, that our study is based on a randomized design, which aims to control for the risk of confounding (both for known and unknown confounders). In our trial, the size of the sample and the randomization process led to successful control of pre-, intra-, and postoperative confounders, as reflected in tables 1 and 2. As for missing data in figure 2, the reason for data loss is mentioned in the Methods section and pertained to early hospital discharge.
Dr. Schläpfer has received a research grant from Roche Diagnostics International Ltd. (Rotkreuz, Switzerland) and a Sedana Medical research grant (Danderyd, Sweden), both not related to this work. The remaining authors declare no competing interests.